They're fantastic for redundant clusters (e.g. something Cassandra-like). Virtualization doesn't help you there -- you actually need the physically separate nod

They're fantastic for redundant clusters (e.g. something Cassandra-like). Virtualization doesn't help you there -- you actually need the physically separate nodes. Keeping power used down directly affects our rack costs, which only provide so much power. We do a lot to stay within that, so we only need 2U (and the included power) instead of 4U or more, since our boards only require 2U. We place all of our clusters nodes in the same physical box, including the networking between them. This allows us to ship the box to a datacenter and all it needs is power, ethernet connections to the outside world, and possibly, the assignment of static IP addresses. Any tech can install the cluster in minutes. We're not using the Atom processors though, we're using the lowest-power Xeon E5 in our current design. The Atom S1200 looks interesting.